Development and performance evaluation of a mixed-sensor system for fine particles and road traffic noise

Abstract

In recent years, the Internet of Things and low-cost sensor technologies have been applied to establish low-cost sensor networks for monitoring single pollutants in the environment. Few studies have developed a mixed-sensor system for simultaneous measurements of particles and noise, but the influences of meteorology and other pollutants are not taken into account. This study aimed to develop a mixed-sensor system for fine particles and noise with low-cost sensor technologies for considering effects of temperature, relative humidity, and carbon dioxide (CO 2 ). This mixed-sensor system was evaluated in the laboratory and field by using regular direct-reading instruments, including a portable dust monitor for fine particles and a class 1 sound level meter for noise measurements. Linear regression models were used to establish the relationships between measured values in the direct-reading instruments and sensor values. The present study established a predictive model of PM 2.5 concentration with a high predictive capacity (R 2 = 0 . 89 ) and good accuracy (bias and precision of 0 . 74 ± 1 . 67 μ g/m 3 ; accuracy of 1 . 82 μ g /m 3 based on relative humidity, CO 2 levels, and PM 2.5 sensor values). A predictive model of noise levels was built with a high predictive capacity (R 2 = 0 . 96 ) and moderate accuracy (bias and precision of 2.92 ± 2.96 dBA; accuracy of 4.16 dBA based on temperature, relative humidity, CO 2 levels, and noise sensor values). The developed predictive models with the high and moderate accuracy for a mixed-sensor system can be applied to monitor PM 2.5 and noise levels simultaneously for exposure assessment in exposure studies. • A mixed-sensor system was developed to monitor PM 2.5 and noise levels. • This low-cost system was validated in the laboratory and field. • Temperature, relative humidity, and CO 2 were considered in the predictive models. • The PM 2.5 model was developed with the high predictive capacity and good accuracy. • The noise model was built with a high predictive capacity and moderate accuracy.